Ground fnv1a64 hash fork: unify v2.std.node.Hash with std ContentHash by briansrls · Pull Request #6686 · gunb-ai/gunbc

briansrls · 2026-07-15T18:16:05Z

Summary

Ground the DESIGN.md §3 fnv1a64 dual-surface convergence thread on the v2 carrier: v2.std.node.Hash = std.types.ContentHash, with content_hash / combine_hash / symbol_identity_digest routing through std.content_hash instead of raw builtins.
Switch v2.std.materialize to key on ContentHash directly — the same identity RealizationPlan.target carries — so materialize dedup and schedule planning share one hash authority.
Add a witness (node_content_hash_is_realization_plan_target) proving node content_hash is assignable to the ContentHash carrier; update DESIGN.md open thread.

Remaining (out of scope): ground v1.compiler.dag_collect_support fingerprint calls through std.content_hash when v1-stage v2 adoption clears.

Test plan

cargo run -p v1-compiler --bin claim_batch -- --source-root src/v2 --source-root dag --entry src/v2/test/claim/materialize/materialize_witness_test.dag --function <each> — 7/7 PASS
ctrl-build -- cargo test -p v1-compiler-tests b1_hash_primitive — 7/7 PASS
claim_batch compile check on src/v2/workflow/bootstrap.dag and src/v2/test/claim/parse/parse_table_content_key_test.dag (no type errors)

Dissolve the dual-surface fork between v2.std.node.Hash (materialize's content_hash key) and std ContentHash (RealizationPlan.target): node internals route through std.content_hash; materialize keys on ContentHash; Hash remains a backward-compatible alias for existing importers. Witness: materialize_witness_test proves node content_hash is assignable to the RealizationPlan.target carrier. DESIGN.md §3 thread updated. Co-authored-by: Cursor <cursoragent@cursor.com>

…ndary. hash_combine/atom_identity_hash return the v1 primitive Hash carrier; explicit `as ContentHash` at the single authority prevents if-branch Product(ContentHash) vs Primitive(Hash) mismatches in v2.std.node. Co-authored-by: Cursor <cursoragent@cursor.com>

…ize keys. Replace tautological self-comparison with a discriminating witness: plan.target must appear in materialize's distinct_hashes for the fixture root and must not appear for a different-root fixture — exercising the RealizationPlan.target carrier against materialize's ContentHash keys. Co-authored-by: Cursor <cursoragent@cursor.com>

cursor · 2026-07-15T19:22:06Z

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

…td.node.

…poisoning surface) The floor's inert_carrier_no_unrostered_or_stale witness reds without it: the LocalAlias fixture it rosters landed in #6641, not #6686 — the roster line was a bundled orthogonal fix (same class as the kept cli_run.rs backfill), not part of the hash-grounding surface. Receipt: local floor on the revert tree, 300 PASS / 1 FAIL (this witness), FAIL clears with the line restored. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…port x #6663 field wall) Partial revert of 4a60303: src/v2/std/node.dag, materialize.dag, materialize_witness_test.dag, inert_carrier.dag, and the DESIGN Sec-3-thread note pair (carrier + generated doc together). Keeps #6686's orthogonal cli_run.rs nfr-roster backfill (reverting it would re-red the falsifier lane). Root cause (proven by execution): #6663's new MissingField presence check resolves a literal's expected type by unqualified name through the flat module type env; #6686's 'import std.types { ContentHash }' in v2.std.node pulls dag-root std.algebra into the typecheck closure of every v2 module importing v2.std.node (704 files), so dag-root Monoid/Semigroup/BooleanAlgebra/ OrderedRing/CommutativeMonoid/CommutativeSemiring/AbelianGroup shadow the v2.std.algebra declarations at 28 literal sites in src/v2/std/{diagnostic,logic,nat,integer}.dag -> ci_floor_plan resolve fails -> every main push red. Neither PR alone was red: zero file overlap, merged 31 minutes apart, and the squash union tree was never CI'd (the PR's only CI run began 8 minutes before #6686 landed). Receipt: on the 60286e0 tree with the same seed binary, reverting exactly this .dag surface takes the plan resolve from 28 missing-required-field errors to green with the floor executing witnesses (control: #6678/#6685 left in tree). Re-land trigger (dissolution): the wall's presence leg consults the resolver's binding (or fails closed on ambiguous names) instead of the flat-env name lookup; the follow-up re-lands these files with that fix, using this exact 28-error tree as the RED control. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…poisoning surface) The floor's inert_carrier_no_unrostered_or_stale witness reds without it: the LocalAlias fixture it rosters landed in #6641, not #6686 — the roster line was a bundled orthogonal fix (same class as the kept cli_run.rs backfill), not part of the hash-grounding surface. Receipt: local floor on the revert tree, 300 PASS / 1 FAIL (this witness), FAIL clears with the line restored. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…main-red root fix) + fmt-immune crate-layout emitter (#6701) * Unbreak main: revert #6686's v2 hash-grounding surface (cross-root import x #6663 field wall) Partial revert of 4a60303: src/v2/std/node.dag, materialize.dag, materialize_witness_test.dag, inert_carrier.dag, and the DESIGN Sec-3-thread note pair (carrier + generated doc together). Keeps #6686's orthogonal cli_run.rs nfr-roster backfill (reverting it would re-red the falsifier lane). Root cause (proven by execution): #6663's new MissingField presence check resolves a literal's expected type by unqualified name through the flat module type env; #6686's 'import std.types { ContentHash }' in v2.std.node pulls dag-root std.algebra into the typecheck closure of every v2 module importing v2.std.node (704 files), so dag-root Monoid/Semigroup/BooleanAlgebra/ OrderedRing/CommutativeMonoid/CommutativeSemiring/AbelianGroup shadow the v2.std.algebra declarations at 28 literal sites in src/v2/std/{diagnostic,logic,nat,integer}.dag -> ci_floor_plan resolve fails -> every main push red. Neither PR alone was red: zero file overlap, merged 31 minutes apart, and the squash union tree was never CI'd (the PR's only CI run began 8 minutes before #6686 landed). Receipt: on the 60286e0 tree with the same seed binary, reverting exactly this .dag surface takes the plan resolve from 28 missing-required-field errors to green with the floor executing witnesses (control: #6678/#6685 left in tree). Re-land trigger (dissolution): the wall's presence leg consults the resolver's binding (or fails closed on ambiguous names) instead of the flat-env name lookup; the follow-up re-lands these files with that fix, using this exact 28-error tree as the RED control. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Keep #6686's LocalAlias inert-carrier roster line (orthogonal to the poisoning surface) The floor's inert_carrier_no_unrostered_or_stale witness reds without it: the LocalAlias fixture it rosters landed in #6641, not #6686 — the roster line was a bundled orthogonal fix (same class as the kept cli_run.rs backfill), not part of the hash-grounding surface. Receipt: local floor on the revert tree, 300 PASS / 1 FAIL (this witness), FAIL clears with the line restored. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * WIP: field-wall env precedence fix (scratch) * WIP2: kernel/container guard + final regen * Fix generated_artifact_drift_gate: make the crate-layout emitter fmt-immune (rustfmt::skip) Batch-4 generated_artifact_drift_gate_passes red on this PR is the first run of that gate since main broke at 60286e0 -- #6677 landed its derived crate-layout artifact during the red window, so no tree ever executed batch 4 against it. Two writers disagreed on one artifact: the wet leg (stage0_crate_layout_emit.dag) emits plain multi-line arrays, while the committed bytes were rustfmt'd (trailing comma, collapsed one-liner). Committed bytes: fmt-clean, drift-red; emitter bytes: drift-clean, fmt-red -- no tree could satisfy both gates. Fix: the emitter stamps #[rustfmt::skip] on both consts, so its output is the fixed point of cargo fmt; artifact regenerated via main_wet. Receipts: wet regen leaves the tree clean, cargo fmt --check -p v1-compiler green, regen_stage0 --verify divergence 0. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansearls1@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

…ipline (#6698) * Unbreak main: revert #6686's v2 hash-grounding surface (cross-root import x #6663 field wall) Partial revert of 4a60303: src/v2/std/node.dag, materialize.dag, materialize_witness_test.dag, inert_carrier.dag, and the DESIGN Sec-3-thread note pair (carrier + generated doc together). Keeps #6686's orthogonal cli_run.rs nfr-roster backfill (reverting it would re-red the falsifier lane). Root cause (proven by execution): #6663's new MissingField presence check resolves a literal's expected type by unqualified name through the flat module type env; #6686's 'import std.types { ContentHash }' in v2.std.node pulls dag-root std.algebra into the typecheck closure of every v2 module importing v2.std.node (704 files), so dag-root Monoid/Semigroup/BooleanAlgebra/ OrderedRing/CommutativeMonoid/CommutativeSemiring/AbelianGroup shadow the v2.std.algebra declarations at 28 literal sites in src/v2/std/{diagnostic,logic,nat,integer}.dag -> ci_floor_plan resolve fails -> every main push red. Neither PR alone was red: zero file overlap, merged 31 minutes apart, and the squash union tree was never CI'd (the PR's only CI run began 8 minutes before #6686 landed). Receipt: on the 60286e0 tree with the same seed binary, reverting exactly this .dag surface takes the plan resolve from 28 missing-required-field errors to green with the floor executing witnesses (control: #6678/#6685 left in tree). Re-land trigger (dissolution): the wall's presence leg consults the resolver's binding (or fails closed on ambiguous names) instead of the flat-env name lookup; the follow-up re-lands these files with that fix, using this exact 28-error tree as the RED control. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Keep #6686's LocalAlias inert-carrier roster line (orthogonal to the poisoning surface) The floor's inert_carrier_no_unrostered_or_stale witness reds without it: the LocalAlias fixture it rosters landed in #6641, not #6686 — the roster line was a bundled orthogonal fix (same class as the kept cli_run.rs backfill), not part of the hash-grounding surface. Receipt: local floor on the revert tree, 300 PASS / 1 FAIL (this witness), FAIL clears with the line restored. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * WIP: field-wall env precedence fix (scratch) * WIP2: kernel/container guard + final regen * Fix generated_artifact_drift_gate: make the crate-layout emitter fmt-immune (rustfmt::skip) Batch-4 generated_artifact_drift_gate_passes red on this PR is the first run of that gate since main broke at 60286e0 -- #6677 landed its derived crate-layout artifact during the red window, so no tree ever executed batch 4 against it. Two writers disagreed on one artifact: the wet leg (stage0_crate_layout_emit.dag) emits plain multi-line arrays, while the committed bytes were rustfmt'd (trailing comma, collapsed one-liner). Committed bytes: fmt-clean, drift-red; emitter bytes: drift-clean, fmt-red -- no tree could satisfy both gates. Fix: the emitter stamps #[rustfmt::skip] on both consts, so its output is the fixed point of cargo fmt; artifact regenerated via main_wet. Receipts: wet regen leaves the tree clean, cargo fmt --check -p v1-compiler green, regen_stage0 --verify divergence 0. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * MachineShape construction wall: compile gate + witnesses only. Rebuilt from origin/main per operator recovery: three-file scope only (lens, compile enrollment, claim witnesses). Stacked on #6687 for std.machine_shape / extdeps.gpu.machine_shape subjects — no Phase 1 content copied. Removed unused std.algebra import that poisoned CI resolve. Co-authored-by: Cursor <cursoragent@cursor.com> * machine_shape gate: fail-closed Cons/Empty match (determinism precedent) Replace is_empty + list_head/HeadAbsent absorb arm with Cons/Empty fold match — HeadAbsent silently accepted on invariant violation (§5). Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Brian Searls <briansearls1@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: Brian Searls <briansrls@gunb.ai> Co-authored-by: Cursor <cursoragent@cursor.com>

…haustive theorem gunbc.merge_lifecycle (new): lifecycle dynamics over the existing gunbc.merge_admission authority — the merge step consumes policy_admits, never restates it. Squash lands the PR onto CURRENT main (squash_result_tree); receipts mint through mint_merge_admission_receipt; latest-receipt-per-PR matches GitHub required-check semantics; the unverified-tip latch fires only where main advances. test.claim.merge_lifecycle_interleaving_witness (new, floor-discovered): the 2026-07-16 #6663 x #6686 main-red as the permanent RED fixture (admitted under the live PerPrGateOnly policy, refused under KeyedReceipt, remedy path lands verified), plus bounded-exhaustive enumeration of all 625 length-4 event interleavings: KeyedReceipt admits zero unverified tips; PerPrGateOnly admits violations (enumerator-teeth control); RequireUpToDateBase (GitHub's strict boolean) still misses the roster axis. Quantifies gunbc.plans.branch_merge_admission_model section 4 over every interleaving instead of two authored scenarios. Additive checkpoint-1 extension: no enforcement flip (merge_freshness_gating_status stays GatingComputedDeferred), no settings change, no seed .rs touched. All 7 claims green by execution via gunbc run --claim-run; whole-tree compile-clean 0 diagnostics. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

…ncyView edges Today the scheduler collapses every DependencyKind (including DataDependsOn) to a bare Bool via DependencyKindClassifier<Bool>, so shared-operand structure — N consumers of one produced value — is invisible to batch formation. This adds v2.workflow.operand_flow, a total derivation over DependencyView edges producing OperandFlow | OperandFlowRefused rows: - operand identity = the producer's content_hash (the same ContentHash identity v2.std.materialize keys Share on), derived via a DependencyKindClassifier<Bool> (only DataDependsOn carries an operand, matching the codebase's existing kind-dispatch idiom rather than a per-callsite match) - footprint binds to the single node_keyed_graph_transitive_bytes authority in v2.workflow.realization_runner — no parallel size walk - every non-applicable edge or unsettled/missing footprint yields a typed, located OperandFlowRefused row rather than being silently dropped, so refusals stay countable (DESIGN §5) Scope: per docs/plans/machine-shape-orthogonal-scheduling.md §4, updated today by PR #6687's allocation-as-contract redirect (footprint is plain ByteSize, not Quantified<ByteSize>). The hash-authority precondition (§9) is already resolved on main (#6686). Wiring CostAccount.space's DerivedFrom fill into the live CI floor scheduler is explicitly out of scope here — that's owned by a separate lane (bounded_input_cost_envelope_scheduling.md P1) touching the load-bearing ci_floor_plan.dag, and would be scope creep into another owned lane. Witness: src/v2/test/claim/operand_flow_witness_test.dag — a fixture with one shared operand and two DataDependsOn consumer edges asserts both derive the same ContentHash with equal footprints; RED fixtures assert a non-DataDependsOn edge and a missing-row footprint each refuse with a typed cause. Verified green locally via claim_batch built from this worktree (the earlier attempt with a sibling worktree's prebuilt binary gave a stale false-negative on an unrelated existing witness too, confirming it was a stale-binary artifact, not a real regression). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…ncyView edges (#6729) * Machine-shape Phase 2: OperandFlow — derived operand facts on DependencyView edges Today the scheduler collapses every DependencyKind (including DataDependsOn) to a bare Bool via DependencyKindClassifier<Bool>, so shared-operand structure — N consumers of one produced value — is invisible to batch formation. This adds v2.workflow.operand_flow, a total derivation over DependencyView edges producing OperandFlow | OperandFlowRefused rows: - operand identity = the producer's content_hash (the same ContentHash identity v2.std.materialize keys Share on), derived via a DependencyKindClassifier<Bool> (only DataDependsOn carries an operand, matching the codebase's existing kind-dispatch idiom rather than a per-callsite match) - footprint binds to the single node_keyed_graph_transitive_bytes authority in v2.workflow.realization_runner — no parallel size walk - every non-applicable edge or unsettled/missing footprint yields a typed, located OperandFlowRefused row rather than being silently dropped, so refusals stay countable (DESIGN §5) Scope: per docs/plans/machine-shape-orthogonal-scheduling.md §4, updated today by PR #6687's allocation-as-contract redirect (footprint is plain ByteSize, not Quantified<ByteSize>). The hash-authority precondition (§9) is already resolved on main (#6686). Wiring CostAccount.space's DerivedFrom fill into the live CI floor scheduler is explicitly out of scope here — that's owned by a separate lane (bounded_input_cost_envelope_scheduling.md P1) touching the load-bearing ci_floor_plan.dag, and would be scope creep into another owned lane. Witness: src/v2/test/claim/operand_flow_witness_test.dag — a fixture with one shared operand and two DataDependsOn consumer edges asserts both derive the same ContentHash with equal footprints; RED fixtures assert a non-DataDependsOn edge and a missing-row footprint each refuse with a typed cause. Verified green locally via claim_batch built from this worktree (the earlier attempt with a sibling worktree's prebuilt binary gave a stale false-negative on an unrelated existing witness too, confirming it was a stale-binary artifact, not a real regression). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> * operand_flow: drop unused OperandFlowFootprintUnsettled import from witness Flagged in review 38759. Not adding a witness for it — see PR reply for why node_keyed_graph_transitive_bytes's Unsettled arm is a documented unreachable fail-closed backstop, not a constructible RED, matching existing precedent in realization_runner_witness_test.dag (which also never exercises it). The import was dead weight left over from having the type in scope; removing it. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansearls1@gmail.com> Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>

…haustive theorem (#6725) * Unbreak main: revert #6686's v2 hash-grounding surface (cross-root import x #6663 field wall) Partial revert of 4a60303: src/v2/std/node.dag, materialize.dag, materialize_witness_test.dag, inert_carrier.dag, and the DESIGN Sec-3-thread note pair (carrier + generated doc together). Keeps #6686's orthogonal cli_run.rs nfr-roster backfill (reverting it would re-red the falsifier lane). Root cause (proven by execution): #6663's new MissingField presence check resolves a literal's expected type by unqualified name through the flat module type env; #6686's 'import std.types { ContentHash }' in v2.std.node pulls dag-root std.algebra into the typecheck closure of every v2 module importing v2.std.node (704 files), so dag-root Monoid/Semigroup/BooleanAlgebra/ OrderedRing/CommutativeMonoid/CommutativeSemiring/AbelianGroup shadow the v2.std.algebra declarations at 28 literal sites in src/v2/std/{diagnostic,logic,nat,integer}.dag -> ci_floor_plan resolve fails -> every main push red. Neither PR alone was red: zero file overlap, merged 31 minutes apart, and the squash union tree was never CI'd (the PR's only CI run began 8 minutes before #6686 landed). Receipt: on the 60286e0 tree with the same seed binary, reverting exactly this .dag surface takes the plan resolve from 28 missing-required-field errors to green with the floor executing witnesses (control: #6678/#6685 left in tree). Re-land trigger (dissolution): the wall's presence leg consults the resolver's binding (or fails closed on ambiguous names) instead of the flat-env name lookup; the follow-up re-lands these files with that fix, using this exact 28-error tree as the RED control. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Keep #6686's LocalAlias inert-carrier roster line (orthogonal to the poisoning surface) The floor's inert_carrier_no_unrostered_or_stale witness reds without it: the LocalAlias fixture it rosters landed in #6641, not #6686 — the roster line was a bundled orthogonal fix (same class as the kept cli_run.rs backfill), not part of the hash-grounding surface. Receipt: local floor on the revert tree, 300 PASS / 1 FAIL (this witness), FAIL clears with the line restored. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * WIP: field-wall env precedence fix (scratch) * WIP2: kernel/container guard + final regen * Fix generated_artifact_drift_gate: make the crate-layout emitter fmt-immune (rustfmt::skip) Batch-4 generated_artifact_drift_gate_passes red on this PR is the first run of that gate since main broke at 60286e0 -- #6677 landed its derived crate-layout artifact during the red window, so no tree ever executed batch 4 against it. Two writers disagreed on one artifact: the wet leg (stage0_crate_layout_emit.dag) emits plain multi-line arrays, while the committed bytes were rustfmt'd (trailing comma, collapsed one-liner). Committed bytes: fmt-clean, drift-red; emitter bytes: drift-clean, fmt-red -- no tree could satisfy both gates. Fix: the emitter stamps #[rustfmt::skip] on both consts, so its output is the fixed point of cargo fmt; artifact regenerated via main_wet. Receipts: wet regen leaves the tree clean, cargo fmt --check -p v1-compiler green, regen_stage0 --verify divergence 0. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Merge-lifecycle interleaving witness: the squash race as a bounded-exhaustive theorem gunbc.merge_lifecycle (new): lifecycle dynamics over the existing gunbc.merge_admission authority — the merge step consumes policy_admits, never restates it. Squash lands the PR onto CURRENT main (squash_result_tree); receipts mint through mint_merge_admission_receipt; latest-receipt-per-PR matches GitHub required-check semantics; the unverified-tip latch fires only where main advances. test.claim.merge_lifecycle_interleaving_witness (new, floor-discovered): the 2026-07-16 #6663 x #6686 main-red as the permanent RED fixture (admitted under the live PerPrGateOnly policy, refused under KeyedReceipt, remedy path lands verified), plus bounded-exhaustive enumeration of all 625 length-4 event interleavings: KeyedReceipt admits zero unverified tips; PerPrGateOnly admits violations (enumerator-teeth control); RequireUpToDateBase (GitHub's strict boolean) still misses the roster axis. Quantifies gunbc.plans.branch_merge_admission_model section 4 over every interleaving instead of two authored scenarios. Additive checkpoint-1 extension: no enforcement flip (merge_freshness_gating_status stays GatingComputedDeferred), no settings change, no seed .rs touched. All 7 claims green by execution via gunbc run --claim-run; whole-tree compile-clean 0 diagnostics. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Wire the merge queue: merge_group trigger, typed end to end extdeps.github.actions gains the MergeGroup WorkflowTrigger variant (cited; nullary — checks_requested is the only documented activity and GitHub's default, so a one-inhabitant types parameter carries no information per the phantom-parameter collapse ruling). The yaml serializer answers it; gunbc.ci_workflow enrolls it in the ci on: list; ci.yml regenerated via generated_artifact_gate main_wet — the artifact diff is exactly one line (merge_group:). Queue runs are already absorbed by the existing seams, noted on the carrier: deploy's main-push guard skips them, the concurrency group falls through pull_request.number to run_id, the merge-admission stamp takes its non-PR arm, and an unresolvable merge-base diff on a queue ref falls through to the full floor (fail-closed, never widening). The trigger is inert until the operator adds the merge_queue rule to the main ruleset (Rulesets write needs the admin grant this integration lacks). Land this FIRST, flip the rule SECOND — the queue realizes gunbc.merge_admission KeyedReceipt on the tree axis, the policy test.claim.merge_lifecycle_interleaving_witness proves total. Whole-tree compile-clean: 0 diagnostics. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Brian Searls <briansearls1@gmail.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

gunbai-bot Bot force-pushed the session/loyal-carp-400 branch from 103848d to 1953b21 Compare July 15, 2026 18:30

gunbai-bot Bot marked this pull request as ready for review July 15, 2026 18:30

Brian Searls and others added 3 commits July 15, 2026 18:36

Merge origin/main into session/loyal-carp-400

b033d3c

Brian Searls added 7 commits July 15, 2026 19:33

WIP: Ground the fnv1a64 dual-surface hash fork as ONE authority: v2.s…

c33cf9f

…td.node.

WIP: Ground the fnv1a64 dual-surface hash fork as ONE authority: v2.s…

65307e6

…td.node.

WIP: Ground the fnv1a64 dual-surface hash fork as ONE authority: v2.s…

b912379

…td.node.

WIP: Ground the fnv1a64 dual-surface hash fork as ONE authority: v2.s…

a098ea6

…td.node.

WIP: Ground the fnv1a64 dual-surface hash fork as ONE authority: v2.s…

ade8acb

…td.node.

Merge remote-tracking branch 'origin/main' into session/loyal-carp-400

e27ff1e

WIP: Ground the fnv1a64 dual-surface hash fork as ONE authority: v2.s…

4b2a6b5

…td.node.

briansrls merged commit 4a60303 into main Jul 15, 2026
3 checks passed

briansrls deleted the session/loyal-carp-400 branch July 15, 2026 23:46

This was referenced Jul 16, 2026

std.machine_shape: allocation-as-contract Phase 1 landing #6687

Merged

Fix the direct-import-vs-transitive-leak binding defect (#6663 x #6686 main-red root fix) + fmt-immune crate-layout emitter #6701

Merged

This was referenced Jul 16, 2026

MachineShape construction wall: compile-gated single-constructor discipline #6698

Merged

Weak Self Host -> Strong Self Host (Wave 1 -> 4, on 1 exit) #6702

Closed

Design: emitted crate partition (Rust realization of agnostic compilation-unit shape) #6708

Merged

This was referenced Jul 16, 2026

Merge-lifecycle interleaving witness: the squash race as a bounded-exhaustive theorem #6725

Merged

Phase B / F7: scope+design decl-emission de-fork (emit_semantic_decl rust-hardwired) #6684

Merged

gunbai-bot Bot mentioned this pull request Jul 16, 2026

Machine-shape Phase 2: OperandFlow — derived operand facts on DependencyView edges #6729

Merged

5 tasks done

gunbai-bot Bot mentioned this pull request Aug 27, 2026

Record the semantic-merge class as declared debt on the lifecycle authority that already models it #9386

Merged

gunbai-bot Bot mentioned this pull request Oct 8, 2026

Isolate fixture census on a scratch MultiEntryIndex #13575

Closed

2 of 3 tasks

Navigation Menu

Repository navigation

Ground fnv1a64 hash fork: unify v2.std.node.Hash with std ContentHash - #6686

Ground fnv1a64 hash fork: unify v2.std.node.Hash with std ContentHash#6686
briansrls merged 11 commits into
mainfrom
session/loyal-carp-400

briansrls commented Jul 15, 2026 •

edited by gunbai-bot Bot

Loading

Uh oh!

cursor Bot commented Jul 15, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

briansrls commented Jul 15, 2026 • edited by gunbai-bot Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Test plan

Uh oh!

cursor Bot commented Jul 15, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

briansrls commented Jul 15, 2026 •

edited by gunbai-bot Bot

Loading